Preparing Photos for AI Image Classification

To ensure reliable and accurate model classification results, it’s important to prepare and capture high-quality photos. The AI model depends on consistent visual input, so attention to photo quality, framing, and orientation directly improves classification performance and reduces the need for manual corrections.

Follow these best practices when capturing or preparing your photos from benthic photo quadrat surveys for classification in MERMAID Collect:

1. Ensure clear, in-focus photos

  • Photos should be well-lit, sharp, and free of motion blur.

  • Avoid photos with poor contrast or heavy shadows, which may obscure benthic features.

  • Use a camera with a resolution of at least 12 megapixels for best results.

  • Photo dimension must be at least 1500 x 1500 pixels.

2. Minimize area outside the quadrat frame

  • Crop your photos to include only the area within the photo quadrat, removing excess background and frame edges where possible.

  • This ensures the AI model classifies only the benthic substrate and avoids mislabeling artifacts outside the sampling area.

  • If your quadrat has a PVC or metal frame, it is acceptable for it to appear in the photo as long as it does not dominate the view.

3. Use a consistent top-down perspective

  • Capture photos perpendicular to the reef surface to minimize distortion, and at a consistent distance from the benthic surface. 

  • Ensure the quadrat is evenly aligned in the photo frame, without tilting or rotation.

4. Maintain consistent scale and coverage

  • Use the same quadrat size and camera height throughout each transect in a survey to keep spatial resolution consistent.

  • Ensure the entire quadrat area is visible and not cropped during capture.

5. Avoid obstructions

  • Remove divers’ fins, measuring tapes, or equipment from the field of view before capturing the photo.

  • Ensure no bubbles, debris, or marine life obstruct the benthic surface.

6. Standardize photo formats

  • Upload photos in JPEG, PJPEG, MPO, or PNG format.

  • Keep filenames clear and consistent (e.g., SiteName_Transect01_Quadrat05.jpg) for easier tracking and management.

Figures showing poor image example on the left and an optimal image example on the right.
Left: poor image for classification. Right: optimal image with clear focus, good lighting, and full coverage of the quadrat.